Assert integrand returns a real value for quadpack routines
Because the quadpack routines call coerce-float-fun to be able to
evaluate the integrand, the realpart is returned by default. Instead,
we bind `*plot-realpart*` to nil so that evaluation can return complex
numbers. Then modify float-integrand-or-lose to check for floats
since it was previously checking for cl:number.
Finally to make this all work, declaim `*plot-realpart*` to be special
because we compile plot.lisp (that defines it) **after** compiling
quadpack.lisp. Reordering causes other problems.
common-lang.mk is included in each lang subdirectory and the Makefile
has a dependency on texi2html.init. Use an absolute path to this
file (in doc/info/texi2html.init). This shows up in building the ja
docs, which complains that texi2html.init doesn't exist which is true
because the ja directory doesn't have a copy.
Bug #4176:
Commit c9dd89bd4 introduced a bug where the third form of calling
fullratsubst no longer works:
fullratsubst(listofeqns,expr);
This commit makes that syntax work. To do this,
1. lratsubst1 iterates *only* over a list of substitutions;
2. a fullratsubst2 now does the iteration over successive applications
of lratsubst1;
3. fullratsubst0,1,2 are called by fullratsubst, depending on the
syntactic form of the inputs;
In addition, 7 tests have been added to tests/rtest6a.mac to exercise
all these types of inputs.
Bug #4177:
The test for lratsubst1 has been replaced by the equivalent test for
lratsubst.
8cf78ac...
by
Robert Dodier <email address hidden>
In APPROX-ALIKE (called to compare results in test cases), require dimensions to match when comparing Lisp arrays.
Use mk:defsys to load lisp files for linearalgebra
This should speed up loading of linearalgebra because we no longer
need to search everywhere for the lisp files. mk:defsys will
load (and compile) the files directly.